hook-tool-output-truncator

This middle-cut truncation strategy reduces token usage by preserving the beginning and end of long tool outputs. It activates only when results exceed configurable limits, ensuring the LLM retains essential context without overwhelming the context window.

Yes, you can set different token limits for grep and webfetch outputs using configurable limits. The system supports a global default threshold alongside per-tool overrides, allowing precise control over token management for each specific tool in your project.

The middle-cut strategy for managing tool output tokens truncates the center section of oversized results while preserving the start and end. This approach ensures the LLM retains essential context from large codebase searches or logs without exceeding token limits.

No, the truncator adds zero overhead to normal tool execution. It features low overhead activation that only triggers when tool outputs exceed configured thresholds, ensuring token management does not impact performance for smaller grep or bash results.

To integrate tool output truncation into your existing hook pipeline, enable the hook in your project and configure per-tool limits. This applies the middle-cut truncation strategy across tools like grep, glob, webfetch, and bash within your large codebase.